hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Yu <yuzhih...@gmail.com>
Subject Re: Tablesplit.getLength returns 0
Date Thu, 06 Feb 2014 22:34:45 GMT
I left some minor comments on github - latest patch looks cleaner.

Nice progress.


On Thu, Feb 6, 2014 at 7:23 AM, Lukas Nalezenec <
lukas.nalezenec@firma.seznam.cz> wrote:

> On 6.2.2014 14:08, Ted Yu wrote:
>
>> Have you considered putting the check inside a static method of
>> RegionSizeCalculator ?
>>
> No, I havent.
> I can do it but we will have to add this check to each use of
> RegionSizeCalculator. There is risk of unnecessary NPEs.
> I prefer readable, simple code over avoiding allocation of few kilobytes
> for hundreds of milliseconds in noncritical code in case somebody turns the
> feature off once a year.
> I will add the static method but i wont use it in the InputFormat.
>
> Btw, maybe the best option is removing the configuration option ;-) - its
> really fast now.
>
>
>> Did you put latest patch into use in a cluster ?
>>
> Yes, just now. it looks pretty good. Its fast and its correlated with
> previous Filesystem algorithm.
>
> https://dl.dropboxusercontent.com/u/77461754/tableSplitReturns0.png
>
> It ignores column families but it works with memStore.
>
> I will write unit tests and then make patch.
>
> Best regards
> Lukas
>
>
>> Thanks
>>
>>
>> On Feb 6, 2014, at 4:48 AM, Lukas Nalezenec <
>> lukas.nalezenec@firma.seznam.cz> wrote:
>>
>>  Right, the instantiation could be skipped but I would prefer keeping it
>>> as it is.
>>> This was my intentional design decision - all functionality should be
>>> encapsulated inside class.
>>> Lukas
>>>
>>> On 3.2.2014 19:37, Ted Yu wrote:
>>>
>>>> The new config is checked inside RegionSizeCalculator ctor.
>>>> Instantiation of RegionSizeCalculator can be skipped if the config says
>>>> disabled, right ?
>>>>
>>>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message